AN INTEGRATED SCIENTOMETRIC AND SNA APPROACH TO EXPLORE THE CLASSICS IN CEM RESEARCH

This study explores the classics that significantly contribute to the research of construction engineering and management (CEM). Previous studies usually simply applied the number of citation to identify the classics, causing some flaws. To overcome the flaws, an advanced approach is developed by integrating scientometric methods (G-index and co-citation analysis) and a social network analysis (SNA) technique (modularity optimization algorithm), thus providing more precise and persuasive results that denote what academic works have made significant inspirations and illuminations on CEM research. This study retrieves 13,273 CEM literature and extracts 336,129 bibliographies from these literature. Based on the G-index, a total of 67 publications are identified as CEM classics. Moreover, this paper measures and maps the structure of the classics by using co-citation analysis and modularity optimization algorithm. The results provide a basic source of academic information representing the foundation of CEM and draw a big picture of CEM to show the underlying associations between the identified classics. This can help researchers recognize the key scientific contributions for improving the academy progress.


Introduction
Classics, defined as academic achievements that are generally recognized by the majority of academic scholars, are supposed to be necessary for an academic field to improve itself (Kuhn, 1962). According to Kuhn's book titled "The structure of scientific revolutions", when recognized as the classics by most of the experts, the academic works actually serve functions such as illuminations and examples including laws, theories, applications and instrumentations to progress and consolidate the scientific field itself (Kuhn, 1962).
After the beginning of the 20th century, the conventional "classics", like Ptolemy's Almagest, Opticks, and Aristotle's Physica, hardly emerged. Correspondingly, the practice of "citation classics" originated in the late 1970s and referred to such academic achievements that had been highly cited in their fields (Garfield, 1977). Classics are viewed as the "gold bullion of science" (Smith, 2007;Stack, 2012), which provide a basic source of academic informa-tion (Brandt et al., 2010) and represent the foundation and intellectual roots of a field and help researchers recognize the key scientific contributions to boost the academy development and progress (Aksnes & Sivertsen, 2004;Cano & Lind, 1991). This paper investigates the academic field of construction engineering and management (CEM). The landscape of the world has been significantly transformed by physical and built environments -buildings and infrastructures, supporting spaces and places in which people live with less effort and more convenience (Gellert & Lynch, 2003;Kennedy, 2015). Our build environment is developed by complex undertakings of construction and engineering processes, mainly delivered as projects with visible products including roads, tunnels, bridges, airports, railroads, facilities, buildings, dams, and utilities (Floricel et al., 2016). Due to their significant impacts on society, environment and economy, managers of projects have a heavy responsibility for the construction projects, not only in a manner that maximizes value and minimizes risks with less time, but also reaching the goal of sustainable development (Flyvbjerg, 2014;Laurance et al., 2014). To meet the requirements from the managemental practice, CEM emerged and developed into a mature field that is essential to provide basic knowledge and to serve the educational purposes (D' Agostino, 2017;Levitt, 2007). After World War II, fundamental research of CEM mainly concerned the topics of planning, designing, construction, and management of buildings and infrastructures (Levitt, 2007), involving a number of researchers from academic communities.
Similar to other engineering disciplines, CEM absorbs and applies knowledge from physics, chemistry, management science, computer science, etc. as the underlying sciences to analyze and optimize the design and construction processes (Levitt, 2007). Thus, CEM is differentiated from management science because it requires the application of engineering principles and skills in organizing and directing people in a complex environment (Morse & Babcock, 2010). These early CEM studies established theoretical foundations, developed tools, provided practical evidence in a localized environment, and improved the academic works through peer and practitioners review processes. In turn, these foundations, tools, and evidence contributed to the developments of thought, beliefs and methodologies in different parts of CEM, such as construction project management, modeling and optimization, construction design and control, building information modeling and so on (Levitt, 2007;Li et al., 2017). This triggers the discussions and concerns about the following questions: Among the plenty of CEM early studies, what academic works had become the classics to guide the research and practice of the different subareas of CEM? How do these classics interrelate with each other?
In order to provide insights into the above questions, this study aims to explore and analyze the classics in CEM by using a comprehensive approach which can provide more precise and robust results. In the field of CEM, several attempts have been made to identify the classics, but they only focus on the subfields (Flyvbjerg & Turner, 2018;Li et al., 2017;Siemiatycki, 2016). The findings in this study can help the scholars to better understand which academic achievements significantly contribute to the CEM research. This study also provides a rational basis for practitioners to refer to, when retrieving and applying the important theories in the CEM area more effectively.
Most of the existing research identified the classics from journal papers, but high-impact publications always appear in references, not the papers (Seglen, 1998;Wu et al., 2020). In addition, the classics were always identified by using an arbitrary value of citation(e.g., top 50 or 100 cited papers) rather than a systemic index (Ho, 2014;Ibrahim et al., 2012;Korevaar & Moed, 1996;Li et al., 2017;Powell, 2016). In the CEM area, previous studies have made some efforts to analyze the classics, but most of them failed to provide robust and convinced outputs as they do not overcome the limitations of conventional citation analysis mentioned above (Flyvbjerg & Turner, 2018;Li et al., 2017;Siemiatycki, 2016). Moreover, such expositions may be unsatisfactory as they merely focused on how the identified classics affect the CEM development, but failed to draw a big picture for CEM that uncovers the underlying associations between different CEM's subareas.
The present study, not merely by applying the citation analysis, but by proposing a new comprehensive approach integrating scientometric methods (G-index and co-citation analysis) and a SNA technique (modularity optimization algorithm) to overcome the significant limitations of citation analysis, explores an advanced approach to identify the classics of CEM, reducing arbitrary and personal bias that always occurred in previous studies and providing precise and robust results. The results can help the academia understand the evolutionary history and the merits of those contributions held to be "classics" in the research field.

Literature review
The widely-accepted method used to explore and identify classics is citation analysis, whose results are typically called citation classics. The citation of an academic achievement refers to the times it appears in the reference lists of the succeeding publications, which is viewed as a straightforward way to assess the achievements in its academic area (Baltussen & Kindler, 2004). Although some scholars hold debates for using the citation as a metric because: 1) the motivation of the citing behavior and the ways to count the citation number may be different, and 2) it does not differentiate the positive and negative credits, it represents a fundamental unit of measurement for assessing scientific performance (Ajiferuke et al., 2010;Brandt et al., 2010;Buschman & Michalek, 2013). Therefore, the socalled "citation classic" papers (Garfield, 1976;Kelly et al., 2010), which are also called highly-cited articles (Aksnes, 2003), top publications (Korevaar & Moed, 1996), classic articles (Long et al., 2014) or classic literatures (Chinn, 2015), have been studied in various academic fields, such as medicine (Kelly et al., 2010), chemical engineering (Ho, 2012), information science (Leydesdorff et al., 2014), knowledge management (Serenko & Dumay, 2015), and education (McLeskey, 2004). However, such methods have rarely applied in the field of CEM to identify and evaluate classical theories, articles or people.
Most previous studies identified the citation classics by setting a threshold value to filter highly-cited papers, such as fixing a minimum number (e.g., 100 or 50) of citations received (Ho, 2014;Ibrahim et al., 2012;Li et al., 2017;Powell, 2016), or a percentage (e.g., the top 0.5 percent highly cited publications) or a number of highly-cited publications (Aminian et al., 2015;Huang et al., 2015;Leydesdorff et al., 2014;Lo et al., 2016;Serenko & Dumay, 2015). In addition, the H-index has been used to objectively offer an unbiased criterion to filter classics (Martinez et al., 2015). H-index is one of the most frequently used scientometric indicators, reflecting the number of a scholar's most cited publications and the number of the citation that the scholar received (Hirsch, 2007). Compared with traditional methods, H-index is regarded as a relatively unbiased approach to assess the scientific performance (Costas & Bordons, 2008;Hirsch, 2005Hirsch, , 2007, but it fails to consider the highly-cited publications and scale-free nature of the distribution of references-citation system (Woeginger, 2008). Another indicator, G-index, in turn, was proposed as an advancement to overcome the drawbacks of H-index, being recognized as an ideal measure of academic productivity (Egghe, 2006). However, regardless of the advantages, the G-index has not been used in identifying the classics.
Although citation analysis is controversial to some degree, given the current stage of academic development of the knowledge base, it represents a basic form of recognition. The main limitations of the previous methods used in the studies of "citation classics" are as follows.
(1) In most studies, journal papers are always the only source items of databases. The "citation classics" are always identified from the retrieved papers in the databases. Therefore only journal papers could be identified as classics, whereas books and reports are excluded from being classics, regardless of their significant role in many research fields. In addition, publications from other fields that might have considerable impacts on the research of the given academic field are not always included as source items in previous studies.
(2) The criteria applied to identify classics is arbitrary to some extent. It is still not convincing, as to the reasons why we must set a threshold value of 100 or 50 on the citation received to identify the classics, or to use the top 0.5 percent. Indeed, every academic field should have a unique number of classics, as each field has a unique development process and citation pattern. Although the H-index does provide an objective criterion for identifying classics, it does not take into account the most highly-cited publications of an academic field (Woeginger, 2008).

Methodology
The proposed approach integrates scientometric and SNA methods. Scientometrics is the study of measuring and analyzing science, technology, and innovation (Lowry et al., 2004). The current study applies two scientometric methods: G-index and co-citation analysis. In addition, an SNA technique, modularity optimization is also embodied in this study. The authors retrieved the academic papers relevant to CEM and extracted all the references cited in those papers. The authors selected 13 prestigious journals in the field of CEM as the data source based on two criteria. First, the journals need to have significant impacts and unanimous recognition in the research community of CEM (Li et al., 2017;Lin & Shen, 2007;Lu et al., 2015;Olawumi et al., 2017;Wing, 1997). For example, Wing (1997) proposed the top CEM journals with highest scores in quality rating, including JCEM, JCEM-ASCE, ECAM, JME, IJPM, AIC and BRI ( Table 1 shows the initials of the journals). Levitt (2007) reported that after the 1980s, computing approached became popular and many research came out in computer-aided journals, like CCE; Li et al. (2017) state that PMJ, IJPM and JME are renowned journals of CEM. Lu et al. (2015) selected JCEM-ASCE, ECAM, JME, IJPM, AIC, CCE and CCIE to conduct their work. Olawumi et al. (2017) found that most BIM research were published in AIC, CCE, CCIE, AEI, JCEM, and JME. Second, the journals need to be included in the Web of Science (WoS) or Scopus databases which are recognized as authoritative (Meho & Yang, 2007). In addition, for the journals of WoS, they have to rank in the top 75% (Q1, Q2 and Q3 in Journal Citation Report) in each of their subject categories. Following these criteria, 13 journals were selected, as shown in Table 1.
Based on the retrieved papers and references, the authors count the local citation count (LCC) of each reference (the occurrence times of each reference in the CEM papers), and G-index is used to filter the widely recognized academic research according to the LCC ( Figure 1 illustrates the analysis procedure). Then the co-citation analysis and SNA are conducted to explore the relations between the classics of CEM. This paper describes the methodologies as follows.

Data collection
Overall, 11 out of the selected 13 journals of CEM were retrieved from WoS. The authors keep the important papers, by selecting the document types of "ARTICLE", "REVIEW", "EDITORIAL MATERIAL", "PROCEEDINGS PAPER" and "LETTER", since they are regarded as important documents types in WoS (van Leeuwen et al., 2013;Waltman, 2016). In addition, different document types have minimal impacts on the size-independent indicators such as H-index and G-index (Waltman, 2016). The papers of the other two journals were searched from the Scopus database, but only "ARTICLE" and "REVIEW" were searched because of the restriction of the database. The retrieval strategy is: "Publication name" = the 13 CEM journals' names. The authors retrieved all the CEM core journals' articles that are available by using the access of the authors' institute(s), and the time span is from 1985 to 2014. As Table 1 shown, four of the CEM core journals started their publications in 1983. Although this may cause the absence of some papers, the impact on the results is minimal. On the one hand, the number of papers not included in our database is rare (only two years are missed in four journals) compared with the papers retrieved from the database, which may have little influence on the data source. On the other hand, the publications that emerged before 1985 may be included as the source data for identifying the classics, because the authors do not explore the classics from papers but the references. In total, 13,273 CEM papers and 336,129 bibliographies cited by these CEM papers were retrieved as the source database for further analysis. On average, each paper of CEM cited 25.32 references. Table 2 shows the number of documents and bibliographies for each document type.

Data collection
The G-index was introduced by Leo Egghe in 2006 to remedy some weaknesses of the popular H-index (Hirsch, 2005). Specifically, the G-index is the largest number that the top g articles cited g 2 citations together (Egghe, 2006). It is formally defined as follows: The G-index was initially designed to capture the overall performance of any information production process (IPP) (Egghe, 2005b). In the general meaning, any social system can be viewed as an IPP which has sources and corresponding items, such as authors (sources) → papers (items), articles → citations, references → citing times, words → occurrence times in a text, web sites → hyperlinks and so on (Egghe, 2009). In this study, the references of the CEM papers are considered as a part of such an IPP (references of the CEM papers → the times they are cited by the CEM papers). All the scientific productivity of IPPs obey the Lotka's law, because any social activity, including research works, is based on the principle "success-breedssuccess" (De Solla Price, 1976;Simon, 1955). So one publication that has been frequently cited is more likely to be cited again than the one that has been less cited. Therefore, for the references of CEM papers, the size-frequency relation between the reference and citation times align with the Lotka's law: According to the axiom above, the relation of g with the total number of sources (= references here) is given as follows: where T is the total number of references, α denotes the degree of scale nature of the system. From Eqn (3), the relation between G-index and α is well described, and g will increase with the degree of the centration, or inequality of the citing times over the references. This relation can also be explained using the Lorenz curve. Through replacing persons with publications and income with received citations, the Lorenz curve (Lorenz, 1905) can be drawn to measure the inequality over references (Egghe, 2005a).
In general, G-index is better than H-index, which is considered as an ideal measure of academic productivity, because the G-index is sensitive to the highly-cited publications and the scale-free nature of the distribution of references-citation system (Woeginger, 2008). Moreover, any research area has a different g which depicts the pattern of each research area. More than showing the size of the classics, the G-index also measures the impact of the core, which is not shared by other impact measures, such as the H-index (Chong et al., 2015).

Co-citation analysis
As a citation-based approach, co-citation analysis aims to explore the leading publications, journals, and scholars that are referenced by the citing academic sources. The fundamental assumption is that the co-citation network performs the essential features and structure of the intellectual base (Chen, 2004). A closer relationship between two publications is established when they are more frequently cited together (McCain, 1990;Ramos-Rodriguez & Ruiz-Navarro, 2004;Small, 1973). Co-citation analysis has been applied to many fields, such as organization studies (Gmur, 2003), human resources management (Fernandez-Alles & Ramos-Rodríguez, 2009), international management (Acedo & Casillas, 2005) and supply chain management (Charvet et al., 2008).
In the document co-citation network, the nodes are the references cited by the publications included in the analysis. This study adopt cosine coefficients to compute the weights of links of the document co-citation network, which was as (Chen, 2004): where cc [i,j] is the co-cited counts of i and j, denoting the times that i and j were simultaneously cited by the same literature. c [i] and c [j] are total citation counts. The cosine coefficients is similar to fractional counting that is proposed by Perianes-Rodriguez et al. (2016). Both methods aim to provide a more objective and comparable measurement of the co-cited units than traditional full counting, making weights of a co-citation link between 0 and 1. This study adopted cosine coefficients because, for constructing the document co-citation analysis, it considers the impacts of all the citing publications, whereas fractional counting focuses on the publications that create the co-citation action -citing more than two references in the analysis (as for the details of cosine coefficients and fractional counting, please see Chen (2004) and Perianes-Rodriguez et al. (2016), respectively).

Modularity optimization
After constructing the co-citation network of CEM classics, we use an SNA technique -modularity optimization to systematically analyze the structure. SNA techniques were frequently adopted to measure the structure of cocitation networks, visually reflecting the intellectual structure of an academic area by conducting advanced algorithms and topological calculations (Pilkington, 2008). SNA techniques can also reveal the high impact units in a network (Wasserman & Faust, 1994) by examining the relational traits of social structures (Scott, 2012). Modularity optimization is an advanced social network technique to detect communities (Newman, 2006). It is defined as (Newman, 2004): With modularity optimization, a network's structure can be identified by dividing the network into several communities based on the nodal linkages. Internal nodes in a community have many more connections than the external ones. Thus, the references in the same community tend to share similar topics, methods, or ideas (Small, 1973). In this study, modularity optimization is applied to identify the structure of the co-citation network in which the measured CEM classics are the nodes. In the co-citation network, a link and its weight denote the similarity of the two linked classics. By looking at the co-citation relationship, tightly-knit groups of CEM classics can be identified.
In summary, the advantages of the approach this study developed are as follows.
(1) Because the authors explore the classics of CEM from the references of journal papers but not the journal papers themselves, we identify what academic works have significantly impacted and illuminated those studies of CEM, rather than what papers of CEM have been highly cited. This means the approach would provide results that are closer to the characteristics of classics from the perspective of philosophy of science: Classics should guide and illuminate subsequent research and practice in the academic field.
(2) The source items are not limited to journal papers.
As mentioned above, the present study identifies the classics from the references. This makes the journal article is no longer the only type of source items for classics. Other publication types such as books and reports that played prominent roles may be included as source items.
(3) The citation number is more reliable. Based on counting the occurrence frequencies of those references, the base criteria for classics is the citing times by the top CEM journal papers, not the citation provided by online databases (i.e., WoS or Scopus). In addition, rather than a concrete number or a percentage, the G-index was employed as a threshold to avoid bias of personal judgment and to make the procedure of identifying the classics more robust and persuasive.

The classics of CEM
As mentioned in the data collection section, a total of 336,129 bibliographies are extracted from the CEM papers. The authors rank the references according to the times they occurred in the bibliographies (LCC) instead of the citation number in online databases (Global citation count, GCC) such as Google Scholar or WoS. Based on the LCC of the bibliographies and their rankings, the G-index of the cited references is calculated using g = 67.
This means the 67 most popular references of 336,129 bibliographies have been totally cited no less than 67 2 = 4489 (4551 > 4489) times by those documents of CEM journals. The 67 publications cited the most are identified as classics of CEM. A supplemental data file delivers more details about the 67 classics, including information on authors, titles, publication sources and years of publication. Regardless of the detailed information, some intriguing observations of the identified classics should be highlighted.
(1) Overall, the most striking finding to emerge from the results is the 67 identified classics of CEM. The number is not set by personal judgment but G-index. Compared to a concrete number or a percentage such as top 100, 50 or 10 the citations they received or top 0.5 percent highly cited publications (Ho, 2014;Ibrahim et al., 2012;Li et al., 2017;Powell, 2016), G-index can reveal the concentration of publication citations within a defined paper pool, taking the scale-free nature of the distribution of references-citation system into consideration (Woeginger, 2008). Moreover, each academic system should have a different value of G-index, and the number of classics in the different academic field should vary, rather than a concrete number. (2) Several classic theories and methods from other academic fields are identified as CEM classics, such as "Case study research: design and methods" (Yin, 1994), "Analytic Hierarchy Process" (Satty, 1980), "Qualitative data analysis: A sourcebook of new methods" (Huberman & Miles, 1994) and so on. As mentioned above, according to the definition, the classics serve to guide and illuminate the following research, and thus might not be limited to the investigated field. Moreover, CEM is widely accepted as an inter-disciplinary field, absorbing and adopting knowledge from other disciplines that may potentially provide classic theories or tools for CEM. Research to date has tended to retrieved academic publications by searching relevant keywords through online databases (Google Scholar and Web of Science) (Li et al., 2017), and set a threshold value by the citation numbers to identify classics. As a result, some publications are not included as source items for classics due to the search strategy, despite their prominent roles in the investigated academic field. Unlike the previous studies, this study identifies classics that are concentrated but not limited to the CEM papers, because the authors identify the classics within the references extracted from the CEM papers. Therefore, this study, in turn, identifies the significant publications from a broader scope that illuminated and impacted on the research of CEM, which is closer to the characteristics of classics from the perspective of philosophy of science. (3) Besides the 45 journal articles, 20 books and 2 reports are identified as classics of CEM (Table 3). Many studies failed to identify books and reports as classics (Ho, 2014), despite their prominent roles in the research progress.

Clustering the classics by co-citation analysis and modularity optimization
The authors build up the co-citation network of by those classics and their co-cited relations. The overview of the co-citation network of CEM is presented in Figure 2. As shown in Figure 2, the nodes are the 67 classics of the CEM, while the links between them denote their cocitation relationships. The size of each node denotes the times they have been cited by the CEM papers, whereas the thickness of links denotes the co-citation strength of the two nodes. The modularity optimization algorithm divides the CEM classics into different groups which can be distinguished by the attached colors.
There are 311 co-citation relationships between 67 references. As shown in Table 4, the average degree is 9.701, which means every publication has 9.701 co-citation links with others, on average. The density is 0.147, which is much higher than many other social networks (Newman, 2004). The higher density indicates a better overall association of the CEM classics.
The authors name each of the split classic group based on manual review of the classics. At least the titles, abstracts and keywords were carefully read. Furthermore, to discern some ambiguous content, we have read the main body of the classics. The details of each group are shown in the following subsections.

Group 1: Project management
The resulting classics include 25 publications in this group, with 8 books and 17 journal papers (Table 5). The majority of the classics are about management issues of projects, such as project success (Cooke-Davies, 2002;Pinto & Slevin, 1988;Shenhar et al., 2001), cost (Flyvbjerg et al., 2002), project organization (Hobday, 2000;Lundin & Söderholm, 1995;Turner & Müller, 2003) and so on. Moreover, from Table 5, the authors can observe that some theories and methods in social science are recognized as  (Numally, 1978), structural equation model (Fornell & Larcker, 1981), case study (Eisenhardt, 1989;Yin, 1994), qualitative data analysis (Miles & Huberman, 1984) and multivariate data analysis (Hair et al., 1998). Those classic theories and methods were introduced and applied for better understanding and mastering many aspects of construction projects, contributing to the reconceptualization of theory and practice of project management, rethinking the organization and management manner of the project, and advocating future research directions of project management.

Group 2: Construction project success and performance
This group contains eight classics that mainly focus on construction project success, process, and performance, grasping the big picture of the construction industry status (Table 6). Therefore, those classics make a significant contribution to CEM research by highlighting the constraints and drawbacks of management practices in construction projects and stressing the need to improve the performance. The classics identified key factors that may lead to construction project success and pointed out how to improve the performance. The most striking works, in this group, are "Constructing the Team" and "Rethinking Construction", with substantially higher LCC than other classics. "Constructing the team" was recognized as the first serious discussion and analysis on the partnering and collaboration in the construction industry, highlighting the significant role of the client to achieve construction project success (Latham, 1994). "Rethinking construction", a report of Construction Task Force, emphasized the need to meet the new requirements of the construction industry (Egan, 1998). Moreover, aiming at performance improvements, this report proposed a framework for possible improvement in the construction industry.

Group 3: Simulation and Building Information Modeling (BIM)
Classics in this group mainly investigate on computer-aided models, including simulation tools (Halpin, 1977;Halpin & Riggs, 1992;Martinez & Ioannou, 1999) and BIM (Table 7). Those studies provided exciting opportunities to advance our knowledge about applying simulation tools on operations and management processes in construction.
Halpin's two works, as the first language of modern construction simulation, systematically proposed a simulation technique -Cyclone. It is one of the general simulation tools, which is frequently used to model repetitive cycles of tasks in construction processes, such as concrete pouring and formwork layout. Another classic is Koo's and Fischer's work, which developed a 4D Computer Aided Design (CAD) model, enabling the visualization of information flows through project organizations. Therefore, 4D CAD model is an alternative to construction project scheduling tools, adding construction processes and temporal information into the 3D model, which could support the collaboration and communication between construction project participants. By integrating with the 3D model, this model can be successfully adopted with intelligible visualizations, Notes: Please see Table 1 for abbreviations used for Journals. Table 5 and thus the managers can easily generate, manipulate and validate the tool through a common medium (Koo & Fischer, 2000). BIM has been widely applied in the construction industry. The classic book "Building product models: Computer environments supporting design and construction" (Eastman, 1999) proposed a common model to integrate several tools in construction project communication. The solution, named "Building Product Modeling" in the first place, is widely recognized as a digital representation of building processes to facilitate exchange and information interoperability.

Group 4: Radio-frequency identifications (RFID)
This group contains four classic papers about the application of Radio-frequency identifications to the construction industry, especially on tracking and locating precast and fabricated storage components (Table 8). As the progress of the semiconductor fabrication technologies and nanotechnologies, the size of the sensors decreased whereas the cost went down dramatically, leading to the broad applications of the communication devices such as RFID to the industry, significantly altering the design, plan and construction processes. As a starting point, Jaselskis and El-Misalami (2003) introduced the RFID technology to the construction industry, in terms of delivery, billing and quality control for concrete. The primary advantage that the RFID has brought to the construction industry is that the process and schedule for concrete can be monitored and notified to job sites and testing labs. Those classics demonstrated how owners, contractors, and suppliers could enhance their operations by using RFID.

Group 5: Applications of Genetic Algorithm (GA) in construction
The classics in this group mainly introduce a useful approach -GA to the construction industry (Table 9). Publications about mathematics methods, such as fuzzy sets (Zadeh, 1965) and genetic algorithm (Golberg, 1989;Holland, 1975), are identified as classics in this group. As a useful approach, the genetic algorithm was used to support the optimization of design configurations for everything from earthmoving fleets to project organizations. For example, it was applied to the construction industry in terms of dealing with time-cost and trade-off problems (Feng et al., 1997), facilitating scheduling (Harris & Ioannou, 1998;Leu & Yang, 1999) and optimization of resource allocation and leveling (Hegazy, 1999). Those findings made important contributions to optimizing construction operations and processes.
Owing to the enormous number of parameters in the large structure, genetic algorithms, neural networks, and dynamic fuzzy wavelet have been developed and applied to improve the size of the structure.

Summary and conclusions
This study explores the classics that significantly contribute to the CEM research. Moreover, the underlying structure of the CEM classics was identified. The results suggest that CEM research was largely driven by classic theories and methods of other fields. For example, some theories of social science and research methods of management constitute the classics of project management (group 1). Similar trends can be observed in radio-frequency identifications (RFID) (group 4), applications of Genetic Algorithm (GA) in construction (group 5) and structural optimization (group 6). Those classics can help the researchers of CEM have better comprehensive management in construction practice, serving as paradigms to investigate the management issues in construction. The results also indicate that the works that conducted depth summarization of confused phenomena and problems in the construction industry and provided novel views to rethink these problems probably become a classic. For example, the classic "Constructing the team" (Latham, 1994) in construction project success and performance (group 2) seriously discussed partnering and collaboration and highlighted the significant role of the client to achieve construction project success.
To effectively identify the classics of CEM, an integrated approach is developed to offset certain drawbacks that always occurred in previous studies. Although it is a tough task to use quantitative techniques to explore the classics due to the ambiguous definition and complex property, the proposed approach provides more persuasive results and can be used by scholars to identify classics in other academic fields of CEM. In specific, this study outlines two major contributions as follow: (1) The aim of the present study has been to identify classics for the CEM field as a whole. As an academic field, CEM emerged and developed for a long time. It is widely recognized as an inter-discipline, absorbing and applying knowledge from other disciplines. In the field of CEM, several attempts have been made to identify the classics, but they only focus on the subfields. Alternatively, this study identifies CEM classics, representing the theoretical and rational foundations of the whole academic field and help researchers recognize the key scientific contributions to stimulate the academic development of CEM.
(2) To identify the classics of CEM with more precise and robust results, the authors develop an integrated approach that overcomes certain drawbacks in previous studies. The existing research has tended to focus on identifying classics from papers, but high-impact publications always appear in the references, not the papers. In addition, the classics were always identified by applying an arbitrary value (e.g., top 50 or 100 cited papers) rather than a systemic index. In the CEM area, previous studies have made some efforts to analyze the classics, but most of them failed to provide robust and convinced outputs as they do not overcome the limitations mentioned above. The approach proposed by this paper, takes the drawbacks mentioned above into consideration, and thus provides persuasive results about the classics for CEM. Moreover, most recent attention has focused on interpreting the classics' impacts on the subsequent research based on manual reviewing and personal judgments. Such expositions may be unsatisfactory as they failed to draw a whole picture for CEM that uncovers the underlying connections between different subareas of CEM. More than merely presenting the classics of CEM, this study uncovers the structure of the classics by using co-citation and modularity optimization algorithm, taking the underlying associations between the classics into consideration. The study is not without limitations. As the definition does not provide explicit statements about the properties of classics, the criteria for the measurement is still lacking. Although citation is widely used to measure the classics for certain fields, there is some controversy in the citation over whether it provides convinced assessment for the scientific performance. In addition, the classics are not prelabeled, and thus the statistical validation for the methods can not be used. The current study offsets this limitation to some extent, by proposing an advanced approach to overcome some of the drawbacks of the citation analysis that has been widely used in previous studies. The authors also advocate further studies to address the limitations. In particular, advanced machine learning and natural language processing techniques may be used to discern the citing behavior by understanding the content where the citation appears.